Building Blocks for a Participatory Web of Things: Devices, Infrastructures, and Programming Frameworks

Transcription

1 Diss. ETH No Building Blocks for a Participatory Web of Things: Devices, Infrastructures, and Programming Frameworks A dissertation submitted to the ETH Zurich for the degree of Doctor of Science Presented by Mihai Vlad Trifa Dipl. Ing. EPFL Born on 12 March 1982, in Oradea, Romania citizen of Bex (VD), Switzerland accepted on the recommendation of Prof. Dr. Friedemann Mattern, examiner Prof. Dr. Cesare Pautasso, co-examiner Prof. Dr. Gero Mühl, co-examiner 2011

2

3 Abstract In less than three decades, computers have evolved from bulky desktop units to tiny, embedded chips that can communicate wirelessly with each other. Battery-powered and stuffed with the latest sensing technologies, these networked embedded devices (NEDs) have been pervading our world and are increasingly implanted in our cars, buildings, and cities. From mobile phones, to surveillance systems, to contact-less payment systems a legion of sentient machines continuously record our behavior. Analyzing the massive amounts of data collected by NEDs would be very valuable in many disciplines. Unfortunately, the usage of NEDs remains cumbersome because a multitude of hardware and software platforms have been developed both in academia and industry and the communication protocols supported by these platforms are often proprietary and incompatible with each other. Without a unified protocol to interconnect the various devices, building and maintaining large-scale applications that integrate NED data and services is an overly complicated process that requires extensive time, expert knowledge, and resources. A common infrastructure that can scale and evolve as new devices are added will be necessary to build a global network of NEDs installed and maintained by different actors. This in turn would lower the access barrier to develop distributed sensing applications (DSA) as devices could be shared and used easily. In this thesis, we address the integration problem in large-scale DSAs and explore how to simplify access to data and services of heterogeneous embedded devices. Nowadays, access to Internet has become cheap and ubiquitous and it is very likely that all sorts of electronic devices will increasingly possess Internet connectivity. Since the mobile Web is rapidly becoming a commodity, this thesis suggests that the existing Web infrastructure and its protocols should be leveraged to exchange data with NEDs. The tremendous success of the World Wide Web as a global platform for easily sharing information and connecting people to computing services attests that simple and loosely

4 coupled approaches offer a high degree of scalability, robustness, and evolvability. This success is due to a large extent to an open and uniform interface which enabled nonexperts to develop interactive hybrid applications rapidly. Unlike most protocols used in embedded computing, Web standards in the Internet are widely used and highly flexible, and in this thesis we suggest that this model would also be beneficial to future embedded computing applications. Our research bridges the fields of Web technologies and embedded sensing into a unified vision called the Web of Things where the Web s well-known standards and tools are leveraged to seamlessly blend NEDs with the existing Web infrastructure. By drawing upon tools and techniques from both domains, we define the fundamental building blocks of the Web of Things as an extension of the current Web paradigms. After evaluating the limitations of current Web technologies with respect to the requirements of NED applications, we propose practical solutions to alleviate these difficulties to enable the development of efficient, event-driven, and scalable DSAs. Finally, we propose an endto-end, fully Web-based framework that fosters fast prototyping of distributed sensing applications that run on top of heterogeneous NEDs. In contrast to existing research in sensor networks, the central question explored in this thesis is how much of the existing Web infrastructure can be reused to accommodate embedded devices. We further examine the common belief that Web standards are inappropriate for building efficient DSAs. Experimental results and prototypes are provided to support the hypothesis that using Web standards for NEDs is possible. Our results further show that the Web is not only a suitable, but actually a desirable medium to build distributed sensing applications that match the requirements for future large-scale sensing systems. We provide a comprehensive conceptual and empirical investigation of the usage of Web standards to exchange information with embedded devices, and the contributions of our work are multiple. First, our results are relevant to the sensor network and pervasive computing communities, as they support the hypothesis that the existing Web ecosystem is sufficient as is to build a new generation of scalable and flexible participatory applications on top of heterogeneous NEDs. Second, the Web community at large can build upon our set of guidelines to push the Web into the physical world by integrating devices in the Web fabric, thus making the idea of a Web API for the real world realistic. Third, we explore the practical usage of Web technologies in various contexts, from smart spaces to smart cities, and show that a fully Web-based infrastructure is an excellent basis to build an ecosystem of reconfigurable cyber-physical systems. Finally, we hope the work presented here will serve as inspiration for future Web developers and sensor network researchers. Bridging the gap between these two worlds will very likely shed light upon an unexplored design space to create more potent solutions for important societal problems, from energy-efficient buildings, to catastrophe detection and response systems, to more livable and enjoyable cities.

5 Résumé En moins de trois décennies, les systèmes embarqués connectés (SEC) sont devenus omniprésents dans notre monde, car de plus en plus implantés dans nos maisons, voitures, ou dans nos villes. Ces ordinateurs miniature peuvent communiquer entre eux en utilisant différents protocoles radio et possèdent toutes sortes de capteurs qui mesurent une myriade de paramètres (température, pression, mouvements, etc.). Des téléphones mobiles, aux systèmes de surveillance (CCTV), en passant par les systèmes de paiement sans contact une légion de machines enregistrent en continu nos comportements. L analyse des quantités massives de données recueillies par ces systèmes serait très utile dans de nombreuses disciplines. Malheureusement, l utilisation de ces machines n est de loin pas à la portée de tout le monde en raison de la variété et de la complexité de plateformes matérielles et logicielles disponibles tant dans le milieu académique qu industriel. De plus les protocoles de communication utilisés par ces plates-formes sont souvent fermés ou sous licence, et surtout incompatibles entre eux. Sans un protocole unique pour interconnecter ces dispositifs hétérogènes, le développement et le maintien d applications à grande échelle qui intègrent les données recueillies par les SEC, resteront des processus complexes qui nécessitent beaucoup de temps, de connaissances, et de ressources. Ces solutions compliquées sont généralement basées sur un couplage étroit entre composants, ce qui entrave fortement la flexibilité et la scalabilité de ces systèmes. De plus, dans la plupart des cas, ces données collectées finissent isolées dans des applications fermées. Une infrastructure commune qui peut s adapter et évoluer en ajoutant de nouveaux SEC sera un élément essentiel pour la réalisation d un réseau mondial de SEC développé et entretenu par différents acteurs. Cette infrastructure faciliterait l usage des SEC et permettrait à un grand nombre d utilisateurs de créer leurs propres applications d analyse de

6 données du monde réel. Dans cette thèse, nous abordons le problème de l intégration à grande échelle de ces systèmes afin de comprendre comment simplifier l accès aux données et aux services offerts par des dispositifs embarqués hétérogènes. Aujourd hui, l accès à Internet est devenu omniprésent et bon marché, de plus il est fortement probable que toutes sortes d appareils électroniques pourront se connecter à Internet. Étant donné que le Web mobile est sur le point de devenir une commodité, nous suggérons que l infrastructure existante ainsi que les protocoles du Web peuvent être mis à profit afin d échanger des données avec les SEC. Le formidable succès du World Wide Web comme plateforme mondiale pour le partage facile d informations et pour se connecter aisément aux services informatiques atteste qu une approche simple offre une robustesse et une scalabilité hors pair. Ce succès est dû dans une large mesure à une interface ouverte et homogène qui a permis à des programmateurs novices de développer facilement des applications interactives hybrides. Contrairement à la plupart des protocoles utilisés dans l informatique embarquée, les standards du Web sont largement utilisés et très flexibles, et dans cette thèse nous suggérons que ce modèle serait également bénéfique pour les futures applications d informatique embarquée. La recherche présentée dans ce document de la combinaison des technologies Web et des systèmes embarqués ont donné naissance à une vision appelée le Web des Objets où les standards et outils du Web sont mis à profit pour intégrer ces SEC au Web de façon transparente. En s appuyant sur les outils de ces deux domaines, nous définissons ici les éléments fondamentaux ainsi que des composants concret qui forment le Web des Objets. Après avoir évalué les limites des technologies Web pour des applications en temps réel, nous proposons des solutions pratiques pour compenser ces limitations afin de permettre la mise en place d applications de captage distribuées, qui sont entièrement basées sur les technologies du Web. Enfin, nous proposons un système complet entièrement basé sur le Web qui favorise le prototypage rapide d applications qui se basent sur ces capteurs sans fil. Nous discutons l idée reçue selon laquelle les technologies Web ne sont pas adaptées aux contraintes des réseaux sans fil et qu il faut choisir des protocoles spécifiquement adaptés et optimisés. Des résultats expérimentaux ainsi que des évaluations de prototypes sont fournis afin de corroborer notre hypothèse : développer des applications de captage distribuées en se basant uniquement sur des technologies Web est non seulement possible, mais serait même souhaitable grâce aux nombreux avantages offerts par les standards Web. Les implications de notre travail sont multiples. En premier lieu, nos résultats sont utiles aux chercheurs en systèmes embarqués car ils bénéficient de nouveaux outils pour développer et tester de nouvelles applications distribuées rapidement. Grâce à la facilité d usage des standards Web ainsi que des évaluations et applications concrètes, nous offrons des bases solides qui confirment l hypothèse que l écosystème du Web permet de construire une nouvelle génération d applications participatives qui utilisent des capteurs

7 hétérogènes. En second lieu, la communauté Web au sens large peut aussi se baser sur nos directives en permettant une meilleure intégration du monde physique dans le Web, rendant ainsi l idée d une API Web pour le monde réel plausible. En troisième lieu, nous explorons l utilisation pratique des technologies Web dans différents contextes, en passant de la domotique aux villes intelligentes, et montrons qu une infrastructure entièrement basée sur le Web serait une excellente base pour un écosystème reconfigurable de systèmes cyber-physiques. Finalement, nous espérons que les travaux présentés ici pourront servir d inspiration pour les développeurs Web et chercheurs en réseaux de capteurs. Combler la séparation entre ces deux mondes pourra très probablement donner lieu à la conception de nouvelles solutions plus efficaces pour faire face aux problèmes majeurs de notre sociétés tels que des bâtiments économes en énergie, les systèmes de détection et de réponse lors de catastrophes naturelles et accidents, ou simplement créer des villes plus efficaces, moins gourmandes en énergie, et plus agréables à vivre.

8

9 Acknowledgments Four years ago, doing a Ph.D. didn t sound *that* hard. After all, many of my friends have done it. What a fool! Looking back at these years, I realize it has been a fantastic journey that has changed me in so many ways. I have learned so many things, met so many people, been to so many places. Even though this thesis is based on the academic convention that prescribes the use of we instead of I, the we used throughout this thesis is justified in many occasions, as the research described in these pages would never have existed without the support, help, and contributions of many people, and I will attempt to thank them here. First and foremost, I would like to thank my advisor Prof. Dr. Friedemann Mattern for being my mentor for all these years. His open mind and trust in me have been detrimental both for my research, but also for my own maturation. He always supported me (and not only financially!) every time I wanted to attend or organize a workshop or a conference, which gave me the chance to circle the globe a few times. His constants yes, I think this is a great idea, go ahead (when many others would have said hmmm, well, not so sure ) have been extremely motivating. Also, I would like to thank my co-advisers, Prof. Dr. Cesare Pautasso and Prof. Dr. Gero Mühl, for their constrictive remarks and useful feedback while writing this thesis. I would also like to thank my awesome colleague and friend Dominique Guinard for all these years of collaboration and fun. He has always been there stuck in mud next to me often deeper yet his everlasting positive attitude and good mood, smile, kindness, and extraterrestrial motivational abilities have been detrimental to help me overcome all the side effects of a PhD, and I cannot thank him enough for all the times he s been there for me. Thanks to all my colleagues and friends from SAP, ETH Zurich, and MIT (there are just way too many to list them all here) for many insightful and constructive discussions: it was great fun working with you.

10 I was blessed to have had the right amount of liberty to create my own research agenda and to interact with many talented and inspiring people throughout the world that have contributed in their own way to shape my research and ideas. I could never thank everyone that inspired me on these few lines (and I am deeply sorry for those I forgot), but I would like to thank in particular Matt Cottam, Adam Dunkels, Hannes Gassert, Fabien Girardin, Nicolas Nova, J-P. Martin-Flatin, Frédéric Montagut, and Erik Wilde for inspiring me and helping me structure my thesis and thoughts. Thanks also the whole wonderful team of the SENSEable City Lab at MIT, particularly Prof. Carlo Ratti, where I was fortunate to spend half a year. Then I would also love to thank all the amazing students I had the chance to supervise: Vlatko Davidovski, Michael Haenni, Andreas Kamilaris, Simon Mayer, Oliver Senn, Thomas Pham, Samuel Wieland. They have contributed to much of the implementation of the various components described in this thesis. I can only hope they learned at least a fraction of all I learned by working with these brilliant folks. I was fortunate enough to have worked as an undergrad with amazing researchers that have taught me the craft of research and hard work (Professors Alcherio Martinoli, Auke Ijspeert, Deborah Estrin, Charles Taylor, and Gordon Cheng), you inspired me so much and I can only hope I have lived up to your expectations. Also thanks to a large bunch of amazing assistants of the DISAL (ex-swis) and BIOROB (ex-birg) you know who you are. Thanks also to all the supporters of the Web of Things for their kind words of encouragement and suggestions. A special thanks goes to our detractors for their critiques (especially the non-constructive ones that gave us even more motivation to work harder!!) and of course to all the readers of our research blog, the WebofThings.com. Finally, thanks to all my friends for providing me moral support in rough times and for being there to cheer and beer me up. A special thanks goes Claudia for being the greatest teaching assistant for my human-to-human interaction graduate class I hope I passed the exam. Above all, I would like to thank my parents and grand-parents for their love and determination to make me a better man, but also for all the sacrifices they made for me. Without their devotion for trying to offer me a better life than they had, I d never have been arrived where I am today. Two thumbs up for all of you!!!

11 Table of contents 1 Introduction Problem Statement Methods Contributions Thesis Outline Background and Motivation Motivation Networked Embedded Devices (NEDs) Wireless Sensor Networks (WSNs) Use Cases and Applications Requirements for an Open and Scalable NED Ecosystem Distributed Systems Architectures On Architectural Styles Object-oriented Architectures (OOA) Services-oriented Architectures (SOA) Resource-oriented Architecture (ROA) Discussion Middleware Solutions Internet as Transport Protocol Internet as Application Protocol Towards the Web of Things Discussion Web-oriented Architectures Web-based Architectures

12 3.1.1 Modern Web Infrastructure Motivation for a Web of Things Representational State-Transfer (REST) Addressable Resources Uniform Interface Representations and Encodings Hypermedia as the Engine of Application State Web-enabled Devices Case Study: Physical Mashups Discussion Web-enabling Embedded Devices Web-enabling Devices Case I: Router-based Enablement IP-based NEDs Embedded Web Servers Evaluation Case II: Proxy-based Enablement Proxy-based Architecture A Smart Proxy for the Web of Things Evaluation Case Study: Energy Monitoring Mashups Discussion Distributed Infrastructures for the Web of Things Structuring the Web of Things Hierarchical Location Modeling Localization InfraWoT - an Infrastructure for the Web of Things Modular Software Architecture Ad-hoc Device and Service Discovery Step I: Network-level Device Discovery Step II: Resource Discovery Querying Service Web-based Ad-hoc Interaction Location-aware Physical Mashups Case Study: Mobile Ambient Meter Discussion Web-based Sensor Data Streams Processing Part I: Event-driven NED Applications Message-oriented Middlewares

13 6.1.2 Web-based Streaming and Messaging Discussion: Messaging Web Messaging System (WMS) Web Messaging Evaluation Discussion: Web Messaging Part II: Stream Processing Engines Query Languages and Data Models Web-based Querying Stream Processing for the Web of Things WISSPR Web-based Stream Processing General Design Principles Web Streams Wisspr System Architecture Module 1: Device Connector (DC) Module 2: Messaging Connector (MC) Module 3: Query Processing Connector (QC) Module 4: Storage Connector (SC) Prototypes and Evaluation Case Study I: Detection Applications Case Study II: Data Collection Applications Discussion Conclusion Future Work Appendices 153 A Microformat Example 153 Bibliography 154 Curriculum Vitae 175

14

15 CHAPTER 1 Introduction Inventions have long since reached their limit, and I see no hope for further development. Julius Sextus Frontinus (Highly regarded engineer in Rome, 1st century A.D.) Two decades ago, a world where everyday objects can sense, analyze, store and exchange information existed only in science-fiction novels. Since then, the colossal progress in embedded systems brought into the world a new class of computing devices networked embedded devices (NEDs) which are tiny computers endowed with wireless communication and various sensors and actuators. NEDs make it possible to build large-scale, distributed sensing applications (DSAs) (e.g., facilities and environmental monitoring) with a much finer spatial and temporal resolution than previously possible. DSAs would be an invaluable help for biology, civil engineering and many other disciplines if only they were simpler to program and to use. Despite the fact that companies building such

16 2 Introduction products are blooming, the majority of academic and industrial research projects have mainly targeted isolated, small-scale, and homogeneous networks customized for rather specific applications. Because most projects rely on customized and often incompatible hardware or software platforms, integration of different systems is very difficult and costly. The vision where people, objects, and environments could be augmented with computational capabilities that provide information and services to assist us in everyday tasks was introduced by Mark Weiser [239] and is commonly referred to as ubiquitous computing. The pervasive availability and access to sensing and computing services embedded in the world around us changes fundamentally the relationship and interaction between humans and the physical world. From traditional desktop computers, digital services become decentralized and can follow us anywhere by allowing continuous interaction with the computing infrastructure via various interfaces. New computing and interaction paradigms such as context-aware computing [106] or natural interaction [231] are therefore required to enable efficient interaction with digitally augmented environments. 1.1 Problem Statement So far, little attention has been paid to the development of simple and functional open systems for embedded devices that do not require experts that re-implement similar functionalities over and over again. A large amount of work is still devoted to low-level programming and to the creation of application-specific user interfaces, which is a waste of resources that could be devoted to the development of the application logic. Despite the increasing popularity of the Maker movement 1 and the flourishing of open source communities, the materialization of Mark Weiser s vision [239] is still impeded by the lack of clear, simple, and interoperable standards that allow embedded systems to interact smoothly at a global scale. Simple tasks such as discovering dynamically devices present in a particular location and interacting with them in an ad-hoc manner, are unnecessarily complex problems that require customized applications. While various protocols, as for example Universal Plug and Play (UPnP) [58], offer powerful mechanisms for discovering and interacting with electronic appliances on a local network, a common ground on which heterogeneous devices using different protocols could interact globally and transparently is still missing. A high-level, uniform, and simple protocol to connect to and interact with heterogeneous devices is necessary to enable applications that can turn raw sensor data into social and economic value. Such a protocol should ideally support both automated data acquisition and direct, ad-hoc interaction with embedded devices, as shown in Figure 1.1. Finally, scalable, distributed applications for sensor data collection, processing, and storage should be easy to implement on top of this framework. 1 See:

17 1.2. Methods Automated data acquisition and control 3 Mobile and browser-based ad-hoc interaction with devices Figure 1.1: Emerging heterogeneous sensor network applications require uniform interfaces for both automatic data acquisition and control, as well as for manual interaction. 1.2 Methods Considering the recent progress in mobile communications, Internet access will very likely soon become a commodity accessible from most mobile devices. According to the IP for Smart Objects Alliance (IPSO) [111], an increasing number of embedded devices will be supporting the IP protocol, so that many physical objects will be connected to the Internet. This convergence of physical computing devices and the Internet provides new design opportunities, as digital communication networks will soon not only contain virtual data (images, text, etc.), but also real-time information from the physical world. Early approaches for developing distributed applications were based on tightly coupled solutions, where each element had full knowledge about the other peers and the functions they offered (so-called RPC architectures). More recently, Web Services standards promoted by organizations such as the World Wide Web Consortium (W3C) or OASIS have become a standard solution for building distributed solutions, and an increasing number of companies are using them for exchanging data across IT applications. Such tightly coupled approaches are suited for interconnecting a few devices in controlled situations with mostly fixed network topologies, but are not flexible enough to handle many resourceconstrained and mobile devices with unknown capabilities in dynamic environments. In the last decade, a newer generation of Web applications commonly referred to as Web 2.0 [193] has become highly popular. In contrast to the early days of the Web, the focus has shifted on the user and user-generated content on the one hand, and on the other hand on a set of technologies (e.g., AJAX, RSS) that support the development of highly

18 4 Introduction interactive interfaces that offer a rich user experience, comparable to common desktop applications. As shown by the high scalability of the Web, simple standards (HTTP/HTML) can support efficient and flexible systems where a large variety of hardware and software platforms coexist and interact smoothly. Open access to data through Web services has enabled information to be reused across independent systems, therefore has lowered the access barrier that allows people to develop their own composite applications. The tremendous success of several Web 2.0 applications (e.g., wikis, open source communities, blogs, and more recently Web mashups) illustrates the benefits of such an open collaboration and information sharing for building large participatory applications. Considering the evolution of the Web and the need for a common technical infrastructure to connect sensing devices and other smart physical objects, we propose in this thesis the creation of the Web of Things as an evolution of the more abstract Internet of Things [182], which extends the Web to encompass embedded computers and sensors using the same principles and standards. Reusing the open and simple standards such as XML, HTTP, or RSS that made the Internet so successful in order to allow physical devices (e.g., sensor and actuator networks, mobile phones, etc.) to offer their services to the wide public through the Web will allow any device to finally speak the same basic language as other resources on the Web. This makes it much easier to integrate physical devices with any other content resource or service on the Web which will enable the Internet to reach out into the real world and sense real-time information about the everything around us. In this thesis, we propose the development of tools which facilitate the publishing and processing of data collected by networked embedded devices, and which promote the open sharing of data and hardware infrastructures by maximizing reuse and scalability, while minimizing coupling between participating devices. People, objects, and environments will then become first class citizens of the World Wide Web, which will make them linkable, discoverable, searchable, just like any other Web resource. To realize the Web of Things vision, we start by dissecting the core architectural principles the modern Web is built upon, and propose a set of patterns for implementing common interaction types and functions commonly used in distributed sensing applications. First, we describe the various methods for connecting NEDs to the Web by reusing the Web model to support the different interaction types required for embedded devices. Afterwards, we design a software framework (gateways) that facilitates the integration of any device into the Web and propose how gateways can be connected together to form a scalable and flexible infrastructure (the backbone) for the Web of Things. We then explore the limitations of the request-response model of HTTP and propose a Web-based messaging mechanism. On top of the infrastructure we implemented WISSPR, a Web-based infrastructure for sensor data stream collection, processing, sharing, and storage that allows to build end-to-end distributed applications for NEDs solely using Web standards.

19 1.3. Contributions Contributions The goal of this thesis is the development of tools that facilitate the publishing and processing of data collected by networked embedded devices, and promoting the sharing of data and hardware infrastructures by maximizing reuse and scalability, while minimizing coupling between participating devices. The contributions of this thesis are multiple. First, we provide a comprehensive conceptual and empirical investigation of the usage of Web standards as an interaction protocol for embedded devices to date. Our results are relevant to the sensor network and pervasive computing communities, as we show that the existing Web standards and tools can be used as is to build a new generation of open scalable and flexible applications on top of heterogeneous networked embedded devices. Second, the World Wide Web community at large can build upon the set of best practices and guidelines to extend the Web as we know it to the real world, by seamlessly integrating devices in the Web fabric, thus making the idea of a Web API for the real-world realistic. By leveraging recent developments in Web technologies such as real-time Web or the semantic Web, we demonstrate that the Web is not only an appropriate but actually a highly desirable medium to build distributed sensing applications that match the requirements for future, large-scale, ad-hoc and heterogeneous sensing systems. In the long run, we expect that extending the existing Internet to naturally integrate physical devices will reshape the Internet into a versatile collection of physical and virtual resources that can be automatically (re)combined at run-time. 1.4 Thesis Outline This thesis is organized as follows: in Chapter 2 we introduce a high-level survey of related work in device integration and data management. Chapter 3 introduces the core architecture of the Web along with the founding principles of the Web of Things. Chapter 4 describes how to Web-enable devices, and in Chapter 5 we explore how to use smart gateways for building scalable infrastructures. In Chapter 6 we present an application framework for sensor data streams management. Finally, we discuss our results and review our contributions in Chapter 7.

20 6 Introduction